Data contracts promise reliable, governed data exchange in decentralized architectures, yet remain difficult to enforce in practice. Although a standard specification is emerging, its low maturity and limited adoption mean data contracts are still authored in various formats. Technical engineers must translate these into enforceable code through tools or manual labor, creating a bottleneck in a growing data landscape.
This research investigates how Large Language Models (LLMs) can help to enforce data contracts. Following a Design Science Research paradigm and grounded in 11 semi-structured interviews with industry experts, it proposes the LLM-Assisted Contract Extraction (LACE) Framework to integrate the data products. Rather than enforcing contracts directly, the framework uses an LLM as a translation layer that converts them into a deterministic, machinereadable intermediary model, expressed as a Resource Description Framework (RDF) knowledge graph against a purpose-built Data Contract Ontology. SHACL validation safeguards its structure, and a provenance lineage model preserves a human-in-the-loop.
The artifact was assessed through an ontology evaluation, an intrinsic evaluation, and an extrinsic evaluation with six practitioners. The results show that structural conformance, rather than content retrieval, is the dominant failure mode, and that example-based prompting is needed for usable, reproducible output. Practitioners valued the framework most for its standardization, locating the main adoption barriers in organizational rather than technical concerns. LLMs, therefore, help to enforce data contracts not by enforcing them directly, but by transforming a manual, error-prone translation step into a validated, repeatable foundation for downstream governance automation.